Add new model: Kimi2-6 #45630
vasqu
merged
68 commits into
huggingface:main
from
zucchini-nlp:kimi
add kimi
7eea6247
modular run conversion
89d373b1
add processing for image for now
44c2063c
video processor, maybe like this
1f940fd9
i think this is defi gemma4 rope
f19aa2dd
add model tester
aacd8086
convert weights
eb3d46b9
fix processing
4a811764
looks fine with dummy weights
0249fa18
chunk/permute rope dynamically
2bc97321
fix repo
fe721522
fix some tests
1e2d3064
.
86029a54
rename to match remote code config
a4bc583d
rename tests as well
0c3267d8
oops
8694a52f
update
ef766b60
some comments
c17f9323
more comments about model code
7a248c21
fast tokenizer loaded, thasnk to moonshot
533a7aae
zucchini-nlp
commented
on 2026-05-11
.
43d9b76f
move my ugly code from cluster to GH, needs mocing to proper conversion
12670305
easier to read, yet not in conversion mapping
5035d6ff
add as conversion
827054f6
small fix
5b024a86
have to go, finish next week!
4084d97f
zucchini-nlp
commented
on 2026-05-12
rebase
b3e826ba
resolve todos
74f4d076
push the ep plan and copy from mixtral for now
db102bfe
Fix compressed-tensors MoE decompression under tensor/expert parallelism
04e1f9ac
merge main
a164282d
fix repo
6850a25a
memory fragmentation
cbe7016a
tokenizer
39a27e07
nit
2c10d232
.
1b12611e
even the official repo has this bug lol
91fa7c61
rebase
775ab74e
deepseek sybc with main, it was refactored
02afd1d0
zucchini-nlp
commented
on 2026-06-30
docs and the last bits
f1f9556a
and this
0245273d
tester uses MoE layers
15bff952
fix reverse rename
d8f45837
some comments
3fc847f7
video tests, flaky for now
dcd985f6
tests!
1f826aca
fix tests
38ba44df
fix repo
2f996e4a
no integration tests for now, too huge ckpts
f8a184b5
maybe? looks like same
e09c272f
!
542b7eb4
more comments
1e953212
run check repo
184aeabf
unused attr, delete
3807e12f
dont import these unless torch is installed
18f9a8e2
skip export tests
d848e09c
vasqu
commented
on 2026-07-03
fix modular nit
cd241c24
aya
b7bd24cb
zucchini-nlp
commented
on 2026-07-03
erm, oke, mvoe it to another folder following fpquant
79596466
IlyasMoutawwakil
commented
on 2026-07-03
use the correct ckpt for tests
8a5c9336
Merge remote-tracking branch 'origin/main' into kimi
29d43b0d
vasqu
commented
on 2026-07-03
Update docs/source/en/model_doc/kimi_k25.md
f9f0d121
Merge branch 'main' into kimi
0a1fcd4d
fix kimi k26 export
5952534f
Merge branch 'kimi' of https://github.com/zucchini-nlp/transformers i…
cb30f093
Revert "fix kimi k26 export"
762aacf9
vasqu
commented
on 2026-07-03
copy from main
ba7f97c7
skip
448c296f
Login to write a write a comment.
Login via GitHub